def test_orbital_field_matrix(self):
ofm_maker = OrbitalFieldMatrix(flatten=False)
ofm = ofm_maker.featurize(self.diamond)[0]
mtarget = np.zeros((32, 32))
mtarget[1][1] = 1.4789015 # 1.3675444
mtarget[1][3] = 1.4789015 # 1.3675444
mtarget[3][1] = 1.4789015 # 1.3675444
mtarget[3][3] = 1.4789015 # 1.3675444 if for a coord# of exactly 4
for i in range(32):
for j in range(32):
if not i in [1, 3] and not j in [1, 3]:
self.assertEqual(ofm[i, j], 0.0)
mtarget = np.matrix(mtarget)
self.assertAlmostEqual(
np.linalg.norm(ofm - mtarget), 0.0, places=4)
ofm_maker = OrbitalFieldMatrix(True, flatten=False)
ofm = ofm_maker.featurize(self.diamond)[0]
mtarget = np.zeros((39, 39))
mtarget[1][1] = 1.4789015
mtarget[1][3] = 1.4789015
mtarget[3][1] = 1.4789015
mtarget[3][3] = 1.4789015
mtarget[1][33] = 1.4789015
mtarget[3][33] = 1.4789015
mtarget[33][1] = 1.4789015
mtarget[33][3] = 1.4789015
mtarget[33][33] = 1.4789015
mtarget = np.matrix(mtarget)
self.assertAlmostEqual(
np.linalg.norm(ofm - mtarget), 0.0, places=4)
ofm_flat = OrbitalFieldMatrix(period_tag=False, flatten=True)
self.assertEqual(len(ofm_flat.feature_labels()), 1024)
ofm_flat = OrbitalFieldMatrix(period_tag=True, flatten=True)
self.assertEqual(len(ofm_flat.feature_labels()), 1521)
ofm_vector = ofm_flat.featurize(self.diamond)
for ix in [40, 42, 72, 118, 120, 150, 1288, 1320]:
self.assertAlmostEqual(ofm_vector[ix], 1.4789015345821415)
def test_orbital_field_matrix(self):
ofm_maker = OrbitalFieldMatrix()
ofm = ofm_maker.featurize(self.diamond)[0]
mtarget = np.zeros((32, 32))
mtarget[1][1] = 1.4789015 # 1.3675444
mtarget[1][3] = 1.4789015 # 1.3675444
mtarget[3][1] = 1.4789015 # 1.3675444
mtarget[3][3] = 1.4789015 # 1.3675444 if for a coord# of exactly 4
for i in range(32):
for j in range(32):
if not i in [1, 3] and not j in [1, 3]:
self.assertEqual(ofm[i, j], 0.0)
mtarget = np.matrix(mtarget)
self.assertAlmostEqual(
np.linalg.norm(ofm - mtarget), 0.0, places=4)
ofm_maker = OrbitalFieldMatrix(True)
ofm = ofm_maker.featurize(self.diamond)[0]
mtarget = np.zeros((39, 39))
mtarget[1][1] = 1.4789015
mtarget[1][3] = 1.4789015
mtarget[3][1] = 1.4789015
mtarget[3][3] = 1.4789015
mtarget[1][33] = 1.4789015
mtarget[3][33] = 1.4789015
mtarget[33][1] = 1.4789015
mtarget[33][3] = 1.4789015
mtarget[33][33] = 1.4789015
mtarget = np.matrix(mtarget)
self.assertAlmostEqual(
np.linalg.norm(ofm - mtarget), 0.0, places=4)
ofm_flat = OrbitalFieldMatrix(period_tag=False, flatten=True)
self.assertEqual(len(ofm_flat.feature_labels()), 1024)
ofm_flat = OrbitalFieldMatrix(period_tag=True, flatten=True)
self.assertEqual(len(ofm_flat.feature_labels()), 1521)
ofm_vector = ofm_flat.featurize(self.diamond)
for ix in [40, 42, 72, 118, 120, 150, 1288, 1320]:
self.assertAlmostEqual(ofm_vector[ix], 1.4789015345821415)
# Featurize dataframe with OFM and time it
start = time.monotonic()
ofm = OrbitalFieldMatrix(period_tag=ROW)
ofm.set_n_jobs(NJOBS)
df = ofm.fit_featurize_dataframe(df, 'structure')
finish = time.monotonic()
print("TIME TO FEATURIZE OFM %f SECONDS" % (finish - start))
print()
# Get OFM descriptor and set up KRR model
krr = KernelRidge()
hpsel = GridSearchCV(krr,
params['orbital field matrix'],
cv=inner_cv,
refit=True)
X = df[ofm.feature_labels()].to_numpy()
# Flatten each OFM to form a vector descriptor
XLIST = []
for i in range(len(X)):
XLIST.append(X[i].flatten())
X = np.array(XLIST)
print(X.shape)
Y = df['formation_energy_per_atom'].to_numpy()
mae, rmse, r2 = 0, 0, 0
# Evaluate OFM
start = time.monotonic()
for train_index, test_index in kf.split(X):
X_train, X_test = X[train_index], X[test_index]
Y_train, Y_test = Y[train_index], Y[test_index]
hpsel.fit(X_train, Y_train)
print("--- OFM PARAM OPT")
print ("ROW ELEMS", ROW)
# Featurize dataframe with OFM and time it
start = time.monotonic()
ofm = OrbitalFieldMatrix(period_tag=ROW)
ofm.set_n_jobs(NJOBS)
df = ofm.fit_featurize_dataframe(df, 'structure')
finish = time.monotonic()
print("TIME TO FEATURIZE OFM %f SECONDS" % (finish-start))
print()
# Get OFM descriptor and set up KRR model
krr = KernelRidge()
hpsel = GridSearchCV(krr, params['orbital field matrix'],
cv=inner_cv, refit=True)
X = df[ofm.feature_labels()].as_matrix()
# Flatten each OFM to form a vector descriptor
XLIST = []
for i in range(len(X)):
XLIST.append(X[i].flatten())
X = np.array(XLIST)
print(X.shape)
Y = df['formation_energy_per_atom'].as_matrix()
mae, rmse, r2 = 0, 0, 0
# Evaluate OFM
start = time.monotonic()
for train_index, test_index in kf.split(X):
X_train, X_test = X[train_index], X[test_index]
Y_train, Y_test = Y[train_index], Y[test_index]
hpsel.fit(X_train, Y_train)
print("--- OFM PARAM OPT")